A change of Ct of 0.1-0.2 corresponds to 7-14% difference in absolute terms, so if you have that degree of variation in your Ct values, this ballpark figure (10%) is exactly what you should see reflected in your absolute quantification.

In other words, it's not a problem.

A variation of around 10% most likely reflects pipetting errors, unless you have very high Ct values (i.e. low number of transcripts), in which case it probably reflects pipetting error AND stochastic partitioning effects.

But in essence, it's fine: this is why everyone does qPCR in triplicate. Pipetting errors can be minimised, but are ultimately unavoidable: doing everything in triplicate and then averaging allows you to more or less account for this, and allows you to see whether your pipetting is up to scratch (and discard values if necessary).

Variation in Ct of 0.1-0.2 isn't bad at all, whereas triplicates with Ct values of 24, 24 and 26 would be a clear example of a mistake somewhere in the process. Just drop the 26, call the mean value 24. Or repeat the run to be sure.

You don't then use this variation to derive any kind of error bar, because it's the same sample in all replicate wells. All it would be is a "error in pipetting", which is not a relevant scientific value. You use replicate runs of the same sample to be confident that your mean value for that sample is accurate, then you work out the means +/- standard deviation/SEM for multiple replicate samples.

Also, I'm unclear as to what part alternative splicing plays in all this: nothing you've mentioned seems relevant to alternative splicing?

I'm quantifying the absolute number of wild type (long form) transcript and short (alternatively spliced) variant of one gene and i'm interested in ratios between them in different animals and tissues.

It becomes relevant with alternative splicing because the amount of small variant (alternatively spliced) can reach 30+ Cts, because there is very small amount of it in total RNA. And I'm not sure what more to do, to reduce this deviation.

I hope you understand my problem :) and thank you for your answer.

Regards.

Ok, Ct values in the 30s is definitely stochastic territory.

Generally speaking, for a PCR reaction with ca. 100% efficiency, a single transcript in a well will give you a Ct of around 35, thus two transcripts: 34, four: 33, and so on. 

So you're likely dealing with...countable numbers of molecules per well, which means errors will be higher just because you won't always get the same number of molecules in each well.

Imagine you're flipping a coin: heads it goes in well 1, tails in well 2. 

If you flip 200000 coins, you can be pretty sure you'll have something very close to 50% in each well. If you flip 5 coins, you know already that you CAN'T have 50% in each well, and you're quite likely have something like 20%/80%.

There is no good solution to this, really. Ideally you'd simply "use more RNA", but if your gene of interest is simply poorly expressed (rather than just coming from a small sample) this isn't an option.

The best thing you can do is use more replicates. If you have a sample with your gene of interest present at 10 molecules per microlitre, triplicate reactions is unlikely to give you an accurate measure, whereas doing it 5 or 6 times is...more likely to give you something close to the true value.

Finally, if you have a very large difference in expression level between your long form and alternative splice form, is it really relevant or appropriate to compare them in absolute terms? If you have 10000 molecules of one, and 10 of the other, doubling your splice variant population from 10 to 20 is a huge increase, but the concommitant reduction in long form (from 10000 to 9990) is not a significant change (and will be entirely lost in background noise anyway, especially once you bring normalisation into play). You don't really gain any information from trying to determine the respective numbers each time.

I would establish a "rough and ready" estimate of the relative levels of transcript (like the 10000:10 ratio above), and accept that "rough and ready" is probably as close as you can get with low transcript numbers anyway, and for all subsequent studies I would simply compare long form values with each other, and alternative form values with each other.